Cryptography provides methods of providing privacy and authenticity for remote communications and data storage. Privacy is achieved by encryption of data, usually using the techniques of symmetric cryptography (so called because the same mathematical key is used to encrypt and decrypt the data). Authenticity is achieved by the functions of user identification, data integrity, and message non-repudiation. These are best achieved via asymmetric (or public-key) cryptography.
In particular, public-key cryptography enables encrypted communication between users that have not previously established a shared secret key between them This is most often done using a combination of symmetric and asymmetric cryptography: public-key techniques are used to establish user identity and a common symmetric key, and a symmetric encryption algorithm is used for the encryption and decryption of the actual messages. The former operation is called key agreement. Prior establishment is necessary in symmetric cryptography, which uses algorithms for which the same key is used to encrypt and decrypt a message. Public-key cryptography, in contrast, is based on key pairs. A key pair consists of a private key and a public key. As the names imply, the private key is kept private by its owner, while the public key is made public (and typically associated to its owner in an authenticated manner). In asymmetric encryption, the encryption step is performed using the public key, and decryption using the private key. Thus the encrypted message can be sent along an insecure channel with the assurance that only the intended recipient can decrypt it. The key agreement can be interactive (e.g., for encrypting a telephone conversation) or non-interactive (e.g., for electronic mail).
The use of cryptographic key pairs was disclosed in U.S. Pat. No. 4,200,770, entitled “CRYPTOGRAPHIC APPARATUS AND METHOD.” U.S. Pat. No. 4,200,770 also disclosed the application of key pairs to the problem of key agreement over an insecure communication channel. The algorithms specified in this U.S. Pat. No. 4,200,700 rely for their security on the difficulty of the mathematical problem of finding a discrete logarithm. U.S. Pat. No. 4,200,770 is hereby incorporated by reference into the specification of the present invention.
In a Diffie-Hellman key exchange, two users (e.g., User A and User B) agree on a common G, g, and q. User A generates, or acquires, a secret number a, where 1<a<q, computes g^a, and sends g^a to User B. User B generates, or acquires, a secret number b, where 1<b<q, computes g^b, and sends g^b to User A. User A then computes (g^b)^a, while User B computes (g^a)^b. Since these two values are mathematically equivalent, the two users are now in possession of the same secret number. A cryptographic key may then be derived from the secret number. The significance of this method is that a private key was established between two users by transmitting information over a public channel (i.e., an adversary sees the information being passed) but without knowing a or b, the key cannot be constructed from the information that is passed over the public channel. If the users keep a and b private and the numbers used to generate the key are large enough so that g^(ab) cannot be mathematically derived from g^a and g^b then only the users know the key. In practice, the most common choice for G is the integers mod n, where n is an integer.
Large keys pose problems not only for the adversary but also for the users. Large keys require large amounts of computational power and require large amounts of time in order to generate and use the key. Cryptographers are always looking for ways to quickly generate the shortest keys possible that meet the cryptographic strength required to protect the encrypted message. The payoff for finding such a method is that cryptography can be done faster, cheaper, and in devices that do not have large amounts of computational power (e.g., hand-held smart-cards).
The choice of the group G is critical in a cryptographic system. The discrete log problem may be more difficult in one group and, therefore, cryptographically stronger than in another group, allowing the use of smaller parameters but maintaining the same level of security. Working with small numbers is easier than working with large numbers. Small numbers allow the cryptographic system to be higher performing (i.e., faster) and requires less storage. So, by choosing the right group, a user may be able to work with smaller numbers, make a faster cryptographic system, and get the same, or better, cryptographic strength than from another cryptographic system that uses larger numbers.
The classical choice for G in a Diffie-Hellman key exchange are integers mod n, where n is an integer as well. In 1985, Victor Miller and Neal Koblitz each suggested choosing G from elliptic curves. It is conjectured that choosing such a G allows the use of much smaller parameters, yet the discrete log problem using these groups is as difficult, or more difficult, than integer-based discrete log problems using larger numbers. This allows the users to generate a key that has the same, or better, cryptographic strength as a key generated from an integer G and is shorter than the integer-based key. Since shorter keys are easier to deal with, a cryptographic system based on a shorter key may be faster, cheaper, and implemented in computationally-restricted devices. So, an elliptic curve Diffie-Hellman key exchange method is an improvement over an integer-based Diffie-Hellman key exchange method.
More precisely, an elliptic curve is defined over a field F. An elliptic curve is the set of all ordered pairs (x,y) that satisfy a particular cubic equation over a field F, where x and y are each members of the field F. Each ordered pair is called a point on the elliptic curve. In addition to these points, there is another point 0 called the point at infinity. The infinity point is the additive identity (i.e., the infinity point plus any other point results in that other point). For cryptographic purposes, elliptic curves are typically chosen with F as the integers mod p for some large prime number p (i.e., Fp) or as the field of 2^m elements (i.e., F2m).
Multiplication or, more precisely, scalar multiplication is the dominant operation in elliptic curve cryptography. The speed at which multiplication can be done determines the performance of a cryptographic method based on an elliptic curve.
Multiplication of a point P on an elliptic curve by an integer k may be realized by a series of additions (i.e., kP=P+P+ . . . +P, where the number of Ps is equal to k). This is very easy to implement in hardware since only an elliptic adder is required, but it is very inefficient. That is, the number of operations is equal to k which may be very large.
The classical approach to elliptic curve multiplication is a double and add approach. For example, if a user wishes to realize kP, where k=25 then 25 is first represented as a binary expansion of 25. That is, 25 is represented as a binary number 11001. Next, P is doubled a number of times equal to the number of bits in the binary expansion minus 1. For ease in generating an equation of the number of operations, the number of doubles is taken as m rather than m−1. The price for simplicity here is being off by 1. In this example, the doubles are 2P, 4P, 8P, and 16P. The doubles correspond to the bit locations in the binary expansion of 25 (i.e., 11001), except for the 1s bit. The doubles that correspond to bit locations that are 1s are then added along with P if the 1s bit is a 1. The number of adds equals the number of 1s in the binary expansion. In this example, there are three additions since there are three 1s in the binary expansion of 25 (i.e., 11001). So, 25P=16P+8P+P.
On average, there are m/2 1s in k. This results in m doubles and m/2 additions for a total of 3m/2 operations. Since the number of bits in k is always less than the value of k, the double and add approach requires fewer operations than does the addition method described above. Therefore, the double and add approach is more efficient (i.e., faster) than the addition approach.
While working on an elliptic curve allows smaller parameters relative to a modular arithmetic based system offering the same security, some of the efficiency advantage of smaller parameters is offset by the added complexity of doing arithmetic on an elliptic curve as opposed to ordinary modular arithmetic. For purposes of determining efficiency, elliptic doubles and elliptic additions are often grouped and considered elliptic operations. To gain even more efficiency advantages by going to elliptic curves, cryptographers seek ways to reduce the cost of an elliptic curve operation, or reduce the number of elliptic operations required. An elliptic curve method that requires fewer operations, or more efficiently executable operations, would result in an increase in the speed, or performance, of any device that implements such a method.
It is no more costly to do elliptic curve subtractions than it is to do elliptic curve additions. Therefore, a doubles and add approach to doing elliptic curve multiplication may be modified to include subtraction where appropriate. There are an infinite number of ways to represent an integer as a signed binary expansion. The negative 1s in a signed binary expansion indicate subtraction in a double-add-subtract method while the positive 1s in the signed binary expansion indicate addition in the double-add-subtract method. For example, 25 may be represented as an unsigned binary number 11001 (i.e., 16+8+1=25) or as one possible signed binary number “1 0-1 0 0 1” (i.e., 32−8+1=25).
In an article entitled “Speeding Up The Computations On An Elliptic Curve Using Addition-Subtraction Chains”, authored by Francois Morain and Jorge Olivos, published in Theoretical Informatics and Applications, Vol. 24, No. 6, 1990, pp. 531-544, the authors disclose an improvement to the double-add-subtract method mentioned above by placing a restriction on the signed binary expansion that results in fewer elliptic additions being required to do an elliptic curve multiplication and, therefore, increase the performance (i.e., speed) of elliptic curve multiplication. Messrs. Morain and Olivos proposed generating a signed binary expansion such that no two adjacent bit locations in the signed binary expansion are non-zero (i.e., two 1s, irrespective of polarity, may not be next to each other). Such a signed binary expansion is called a non-adjacent form (NAF) of a signed binary expansion. It has been shown that a NAF signed binary expansion is unique (i.e., each integer has only one NAF signed binary expansion) and contains the minimum number of 1s, irrespective of polarity. By minimizing the 1s, the number of additions is minimized. The improvement proposed by Messrs. Morain and Olivos still requires m doubles but only requires an average of m/3 additions for a total of 4m/3 elliptic curve operations. This is less than the 3m/2 elliptic curve operations required by the classical double and add method described above.
The Diffie-Hellman key exchange algorithm provides confidentiality but not user identification. In other words, the algorithm does not ensure that user A is really communicating with user B, and vice versa. When using Diffie-Hellman, therefore, some additional algorithm must be carried out to identify the users to each other. A more efficient approach is to combine the key agreement and user identification functionalities in a single algorithm. Such an algorithm is called authenticated key agreement.
The typical elliptic-curve based authenticated key agreement method works as follows. The users have agreed in advance on a field F, a curve E, and a base point P of order q. User A has generated a private key wa and a corresponding public key Wa=waP and similarly user B has generated a private key wb and a corresponding public key Wb=wbP. User A generates a private key ra and a corresponding public key Ra=raP and sends Wa and Ra to user B. Similarly, user B generates a private key rb and a corresponding public key Rb=rbP and sends Wb and Rb to user A. User A now combines the values wa, ra, Wb, and Rb in a certain way to obtain a number ca, and also combines the values wa, ra, Wb, and Rb in a second way to obtain a number ga, Similarly, user B combines the values wb, rb, Wa, and Ra in a certain way to obtain a number cb, and also combines the values wb, rb, Wa, and Ra in a second way to obtain a number gb. This is done in such a way that cawb+garb and cbwa+gbra are equal modulo q. User A now computes the shared secret value by evaluating caWb+gaRb, and User B computes the same value by evaluating cbWa+gbRa. An example of this kind of protocol is the MQV algorithm which is described in a paper by L. Law et al, entitled “An Efficient Protocol for Authenticated Key Agreement,” published as Technical Report CORR 98-05, Dept. of C&O, University of Waterloo, Canada, in 1998.
The most expensive part of the above process is that of computing the expressions cW+gR, where c and g are integers and W and R are points on the curve. Thus, it particularly important to optimize the efficiency of this operation.
The most straightforward way to evaluate cW+gR is to evaluate cW and gR separately and add the results. However, it turns out to be more efficient to evaluate the entire expression at once. Such a method is commonly referred to as twin multiplication.
The simplest twin multiplication method was first disclosed by E. G. Straus and later rediscovered by A. Shamir and disclosed in an article by T. ElGamal entitled “A Public Key Cryptosystem and a Signature Scheme Based on Discrete Logarithms,” IEEE Transactions On Information Theory, Vol. IT-31, No. 4, Jul. 1985. The method is based on the binary method which uses an ordinary binary expansion of c and g. Therefore, the Straus-Shamir method is a double/add method for twin multiplication. It is more efficient to use the analogous method that works with signed binary expansions; this is called the double-add-subtract method for twin multiplication. Like the binary method, the double-add-subtract method for twin multiplication works in a general group setting.